Getting Started

• The tutorial can be found at bit.ly/ml-senior-leaders
• We’ll be spending most of today looking at the Kaggle Titanic Dataset
• So lets take a look at our data!

Practical 1 - Exploring data

Linear Models

But we’re not restricted to one factor!

What are we doing?

• \(y_i = \beta_0 + \kappa T_i + \beta_1 X_{1i} + ... +\beta_k X_{ki} + u_i\)
• Really, we’re just fitting a line
• But that line can get super, super squiggly
• Machine learning is just using compute to make the best multidimensional squiggle

Practical 2 - Linear Models

Text and NLP

• We haven’t used the name category at all
• How would you extract value from it?

Image Recognition

• Thinking beyond tabular data, how could what you’ve learnt be applied to images?

But those aren’t straight lines…

So how do we fit the super squiggly line?

Testing and Validating

• Traditional statistics often predicted “in sample”
• Instead, we use test data (which is thankfully a lot less complicated)
• How does that affect our model?

Practical 3 - Metrics

Meaningful testing and meaningful metrics

• So think really hard about:
  • whether your test set is meaningful
  • what baseline performance is
• The magic of ML is that we don’t need to know what our data looks like

As they say on the internet…

Metrics that matter

Linear Models Recap

• Linear models are generally excellent places to start
• Use them to think about your data: how it’s built, how it’s connected, and what you’re aiming to achieve
• With a bit of work, these can perform shockingly well
• but now it’s time to go deeper

The ML Ecosystem

• We rarely have to implement ML from scratch in modern data science
• In fact, it’s often conveniently bundled!
• Which leads to daunting documentation!

So how easy is it?

from sklearn.datasets import load_iris
from sklearn.model_selection import train_test_split
from sklearn.naive_bayes import GaussianNB

X, y = load_iris(return_X_y=True)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=0)
gnb = GaussianNB()
gnb.fit(X_train, y_train)
y_pred = gnb.predict(X_test)

Practical 4 - Machine Learning

Gradient, Loss and Boosting - When lines get squiggly

• Neural networks approaches rely on gradient descent to maximise how well they fit
• Gradient boosters rely on combining (or boosting models)
• Combining your forecasts can be hugely effective

Combining Models

• Boosting is only one approach
• Just like Random Forests, the best models are often combinations
• Lets implement it!

Risks and Benefits to Ensemble

• Has your model performed better?
• What is the implication for fit and variance?

Class Imbalance

• How imbalanced is our model?
• what impact do you think this has in reality?
• There are approaches to tackle this, but consider them early!

Evidence House and the AI Digest